Qualcomm AI Engine Direct - Reduce redundant observers

[winskuo-quic]

Summary

• Current observer flow checks min/max/axis values to determine whether 2 observer is same. If same, it would only keep 1 observer. Providing these values helps observers to properly compare with other nodes observer, which helps reducing the number of observers, which further helps reducing time to quantize the model.

Before:

---

After:

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/6351

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures

As of commit 66c3e44 with merge base f7e26d7 ():

NEW FAILURES - The following jobs have failed:

• pull / test-binary-size-linux-gcc / linux-job (gh)
  RuntimeError: Command docker exec -t 2faba9fe80a8bee8d16129f0ca36a9513b804626405921577955092fc9a1f52a /exec failed with exit code 1
• pull / unittest-arm / linux-job (gh)
  RuntimeError: Command docker exec -t 4890673c7d2edb7af556617151b6d6c90e49a7c81896eaad373e5c6b218da57a /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[winskuo-quic]

Hi @cccclai,
As discussed previously, we want to reduce the numbers of observers to reduce AOT time.
This PR should be able to reduce some observers.
Please have a look.
Thanks

[cccclai]

Looks good, thank you

[facebook-github-bot]

@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[cccclai]

Hi sorry I miss merging this PR, mind rebasing again?

[winskuo-quic]

Hi sorry I miss merging this PR, mind rebasing again?

All good!
I have just rebased.
Thanks

[facebook-github-bot]

@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[digantdesai]

Seems like there are still merge conflicts, might need another rebase?